Exploring Resiliency to Natural Image Corruptions in Deep Learning using Design Diversity